Crimson AI NewsA CrimsonLingua Network service
EN ع
← Back to news
Research paper Hugging Face

RibAssist 3D: Selective Biplanar Rib-Fracture Detection and 3D Localization from CT Projections

AI By Crimson AI Hugging Face Papers 16 August 2026 · 00:00 20 views
Share: X Telegram

A new open-source research prototype, RibAssist 3D, pairs fractures across orthogonal CT projections for accurate 3D localization, but identifies cross-view correspondence as the main bottleneck, achieving a controlled yield with high precision.

RibAssist 3D: Selective Biplanar Rib-Fracture Detection and 3D Localization from CT Projections

Key points

Rib fractures are a common and time-consuming challenge in computed tomography (CT) interpretation. A new research paper introduces RibAssist 3D, an open-source prototype that detects rib fractures in two orthogonal CT-derived projections (anteroposterior and lateral), pairs them across views, and triangulates them into 3D points—while explicitly modeling uncertainty and abstaining when cross-view evidence is insufficient.

The study demonstrates that with correct correspondence, localization is highly accurate: median error of 4.0 mm, with 88% of points within 10 mm and 93.6% rib-exact. On a sealed 55-case cohort, 61.1% of fractures were visible in both views, and a correct pair existed in the candidate graph for 58.4% of fractures. However, the binding limitation is not geometry but confidence-limited cross-view correspondence.

A controlled factorial experiment attributed operational gains to lateral-detector quality rather than matching methods; retraining the lateral detector produced the first nonzero controlled-budget reconstructions. Under a conservative commitment policy, a pre-specified sealed pass promoted 15 of 601 fractures to correct 3D localizations at 0.436 false points per case (2.50% end-to-end commitment yield), with committed points accurate to a median of 1.49 mm and 93% rib-exact.

The low yield is a deliberate consequence of confidence-gated abstention, not a failure of geometry or detection. The authors argue that the framework establishes a reproducible method for selective 3D localization and identifies cross-view correspondence as the dominant operational bottleneck. Code and demo are publicly available.

MetricValue
Median localization error (with correct correspondence)4.0 mm
Points within 10 mm88%
Rib-exact (with correct correspondence)93.6%
Dual-view availability (sealed cohort)61.1%
Correct pair in candidate graph58.4%
Fractures promoted to correct 3D (sealed pass)15 of 601
False points per case0.436
End-to-end commitment yield2.50%
Median error of committed points1.49 mm
Rib-exact of committed points93%
Source
Hugging Face · Hugging Face Papers
Related news
Research paper
Hugging Face 29 Aug 2026

Hugging Face Audit: 110 of 124 AI Evaluations Fail to Support Their Claims

A new commit-bound census of 124 Inspect Evals units reveals that 110 stop before deterministic inference due to missing historica...

4
Research paper
Hugging Face 29 Aug 2026

Aphanta: New Framework Diagnoses When Image Editing Boosts Multimodal Reasoning

Hugging Face researchers introduce Aphanta, a diagnostic framework that evaluates when image-editing intermediates improve multimo...

5
Research paper
Hugging Face 29 Aug 2026

EditaLive! Enables Real-Time Character Video Editing for Live Streaming

Hugging Face researchers introduce EditaLive, a framework for real-time human-centric video editing in live streams, achieving sta...

4